Gpnmb and Spp1 mark a conserved macrophage injury response masking fibrosis-specific programming in the lung

Abstract

Macrophages are required for healthy repair of the lungs following injury, but they are also implicated in driving dysregulated repair with fibrosis. How these 2 distinct outcomes of lung injury are mediated by different macrophage subsets is unknown. To assess this, single-cell RNA-Seq was performed on lung macrophages isolated from mice treated with LPS or bleomycin. Macrophages were categorized based on anatomic location (airspace versus interstitium), developmental origin (embryonic versus recruited monocyte derived), time after inflammatory challenge, and injury model. Analysis of the integrated dataset revealed that macrophage subset clustering was driven by macrophage origin and tissue compartment rather than injury model. Gpnmb-expressing recruited macrophages that were enriched for genes typically associated with fibrosis were present in both injury models. Analogous GPNMB-expressing macrophages were identified in datasets from both fibrotic and nonfibrotic lung disease in humans. We conclude that this subset represents a conserved response to tissue injury and is not sufficient to drive fibrosis. Beyond this conserved response, we identified that recruited macrophages failed to gain resident-like programming during fibrotic repair. Overall, fibrotic versus nonfibrotic tissue repair is dictated by dynamic shifts in macrophage subset programming and persistence of recruited macrophages.

Keywords: Fibrosis; Immunology; Inflammation; Macrophages.

Figure 1. Macrophage isolation and scRNA-Seq following bleomycin and LPS-induced lung injury.

(A) Timeline of interventions in experimental models. Tamoxifen was given to label resident interstitial macrophages followed by a 4-week wait period to allow clearance of labeled monocytes from the circulation. Mice were treated with i.t. LPS (20 μg) or bleomycin (1.5 U/kg) and were euthanized at the times indicated. Anti-CD45 antibody was instilled into the lungs immediately after euthanasia. (B) Four groups of cells were isolated from digested lung tissues using FACS. These included airspace macrophages (AMs), resident IMs (RIM), recruited IMs (RecIM), and general leukocytes (CD45⁺). AMs versus IMs were distinguished based on i.t. administered anti-CD45 labeling. Resident versus recruited IMs were distinguished by tdTomato expression. (C) Fully integrated UMAP including all sorted cells from day 0; LPS days 3, 6, and 15; and bleomycin days 3, 7, and 14. Clusters were manually annotated based on cluster-defining genes. Macrophage clusters are outlined in the dashed black line. (D) UMAP split to show homeostasis, LPS, and bleomycin samples. (E) Feature plot of the log₂ minimum expression of mouse panmacrophage markers Fcgr1, C5ar1, CD68, Mrc1, and Mertk. (F) Average expression and percent of cells expressing panmacrophage marker genes in each cluster. Macrophage clusters in the dashed black box were used in subsequent analyses.

Figure 2. Macrophage clusters correspond to compartment and origin.

(A) Macrophage-specific UMAP generated by reclustering macrophage clusters from the global UMAP. (B) Relative contribution of alveolar macrophages (AM) versus interstitial macrophages (IM) to each cluster, based on sort (AM versus IM) from which the cells were derived. (C) Feature plots of the AM sort from all time points and models showing expression of Siglecf and Itgam, and a UMAP showing the AM sorts exclusively from homeostasis samples. (D) Fraction of cells derived from RIM versus RecIM sorts for IM clusters. (E) Feature plots of IMs from all time points and models showing expression of Folr2 and Ccr2, and a UMAP showing IMs derived exclusively from homeostasis samples. (F) Dot plot of selected marker gene expression for the 7 macrophage clusters. (G) Cytoscape network visualization of GO pathways enriched in the marker lists for each cluster. Nodes correspond to individual pathways, and edges connect nodes with shared genes. Edge color indicates the cluster that is most highly enriched for that pathway; however, multiple clusters may be enriched for a given pathway. (H) DecoupleR transcription factor inference heatmap showing activity scores of the top 25 most variable transcription factors between the 7 clusters.

Figure 3. Macrophage cluster sizes vary over time and between models.

(A) Macrophage UMAP split by model and time point. Vertical bars show the percent contribution of each cluster in the CD45⁺ sort to the macrophage pool for each model and time point. (B) Total macrophage numbers in the lungs from mice treated with i.t. LPS or bleomycin. (C–I) Estimated cell numbers for clusters in LPS (purple) versus bleomycin (green). (J) Heatmap of cell counts for each cluster at each model time point. Line graphs in B–I show mean ± SEM.

Figure 4. Gpnmb RecAM macrophages are found in fibrotic and nonfibrotic lung disease in mice and humans.

(A) Dot plot of expression of fibrosis-associated macrophage marker genes (fibrotic features) in macrophage clusters, scaled across rows. For each cluster, enrichment scores were calculated comparing the overlap of conserved cluster markers and fibrotic features to the expected number of overlapping genes by chance. These are represented in the heatmap. Data are from LPS and bleomycin models and include all time points. (B) Violin plot of fibrotic features gene set scores in Gpnmb RecAM from each model and time point. (C) Violin plot of fibrotic features gene set scores in macrophage clusters from homeostasis. (D–G) Heatmap of enrichment scores of the Gpnmb RecAM gene set and the fibrotic features gene set in macrophage subset marker lists from: human IPF lungs (D), healthy human bronchoalveolar lavage (E), human severe COVID-19 (F), allergic asthmatics and nonasthmatic controls (G), nonfibrotic murine models of pulmonary pneumocystis infection (H), and skin wounds in healthy and diabetic mice (I). Hypergeometric tests in B and C. *P < 0.05, ***P < 0.0005, *****P < 0.000005.

Figure 5. Differential gene expression in Gpnmb RecAM from LPS versus bleomycin.

(A) Number of DEGs between LPS and bleomycin within each cluster at corresponding time points (FDR < 0.05). (B) Sankey plot of DEGs from Gpnmb RecAM at each comparison time point. Significance is based on logFC > 0.5 and FDR < 0.05. (C) Volcano plot of differentially expressed genes (DEGs) for Gpnmb RecAM LPS day 6 versus bleomycin day 7. (D) Cytoscape network visualization of Gpnmb RecAM pathway analysis performed on DEGS up in bleomycin day 7 versus up in LPS day 6. Circles represent enriched pathways and are clustered by similarity. Lines connecting the circles represent genes that overlap between pathways. Circle halves are shaded from grey (no pathway enrichment) to red (high enrichment) with left sides representing LPS and right sides representing bleomycin. Line colors mark the primary group showing significant pathway enrichment, either bleomycin (green) or LPS (purple). (E) Heatmap of transcription factor activity inference scores that were significantly different between Gpnmb RecAM from bleomycin day 7 versus LPS day 6. Only the top 8 transcription factors with the lowest adjusted P values for each injury model are shown. (F) Top 10 GO pathways enriched in the 230 DEGs upregulated in Gpnmb RecAM from bleomycin at both day 7 and day 14 compared with LPS day 6 and 15. Adjusted P < 0.05.